infer_serverless.py

import torch, os, traceback, sys, warnings, shutil, numpy as np
import gradio as gr
import librosa
import asyncio
import rarfile
import edge_tts
import yt_dlp
import ffmpeg
import gdown
import subprocess
import wave
import soundfile as sf
from scipy.io import wavfile
from datetime import datetime
from urllib.parse import urlparse
from mega import Mega

import base64
import tempfile
import threading
import hashlib
import os
import werkzeug
from pydub import AudioSegment
import uuid
from threading import Semaphore
from threading import Lock
from multiprocessing import Process, SimpleQueue, set_start_method,get_context
from queue import Empty
from pydub import AudioSegment

import io
import runpod
import boto3





now_dir = os.getcwd()
cpt={}
tmp = os.path.join(now_dir, "TEMP")
shutil.rmtree(tmp, ignore_errors=True)
os.makedirs(tmp, exist_ok=True)
os.environ["TEMP"] = tmp
split_model="htdemucs"
convert_voice_lock = Lock()
# Define the maximum number of concurrent requests
MAX_CONCURRENT_REQUESTS = 2  # Adjust this number as needed

# Initialize the semaphore with the maximum number of concurrent requests
request_semaphore = Semaphore(MAX_CONCURRENT_REQUESTS)

task_status_tracker = {}
os.environ["OAUTHLIB_INSECURE_TRANSPORT"] = "1"  # ONLY FOR TESTING, REMOVE IN PRODUCTION
os.environ["OAUTHLIB_RELAX_TOKEN_SCOPE"] = "1"
ACCESS_ID = os.getenv('ACCESS_ID', '')
SECRET_KEY = os.getenv('SECRET_KEY', '')

#set_start_method('spawn', force=True)
from lib.infer_pack.models import (
    SynthesizerTrnMs256NSFsid,
    SynthesizerTrnMs256NSFsid_nono,
    SynthesizerTrnMs768NSFsid,
    SynthesizerTrnMs768NSFsid_nono,
)
from fairseq import checkpoint_utils
from vc_infer_pipeline import VC
from config import Config
config = Config()

tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]

hubert_model = None

f0method_mode = ["pm", "harvest", "crepe"]
f0method_info = "PM is fast, Harvest is good but extremely slow, and Crepe effect is good but requires GPU (Default: PM)"






if os.path.isfile("rmvpe.pt"):
    f0method_mode.insert(2, "rmvpe")
    f0method_info = "PM is fast, Harvest is good but extremely slow, Rvmpe is alternative to harvest (might be better), and Crepe effect is good but requires GPU (Default: PM)"




def load_hubert():
    global hubert_model
    models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
        ["hubert_base.pt"],
        suffix="",
    )
    hubert_model = models[0]
    hubert_model = hubert_model.to(config.device)
    if config.is_half:
        hubert_model = hubert_model.half()
    else:
        hubert_model = hubert_model.float()
    hubert_model.eval()

load_hubert()

weight_root = "weights"
index_root = "weights/index"
weights_model = []
weights_index = []
for _, _, model_files in os.walk(weight_root):
    for file in model_files:
        if file.endswith(".pth"):
            weights_model.append(file)
for _, _, index_files in os.walk(index_root):
    for file in index_files:
        if file.endswith('.index') and "trained" not in file:
            weights_index.append(os.path.join(index_root, file))

def check_models():
    weights_model = []
    weights_index = []
    for _, _, model_files in os.walk(weight_root):
        for file in model_files:
            if file.endswith(".pth"):
                weights_model.append(file)
    for _, _, index_files in os.walk(index_root):
        for file in index_files:
            if file.endswith('.index') and "trained" not in file:
                weights_index.append(os.path.join(index_root, file))
    return (
        gr.Dropdown.update(choices=sorted(weights_model), value=weights_model[0]),
        gr.Dropdown.update(choices=sorted(weights_index))
    )

def clean():
    return (
        gr.Dropdown.update(value=""),
        gr.Slider.update(visible=False)
    )
# Function to delete files
def cleanup_files(file_paths):
    for path in file_paths:
        try:
            os.remove(path)
            print(f"Deleted {path}")
        except Exception as e:
            print(f"Error deleting {path}: {e}")





def upload_file(local_file_path,bucket_name):
    # Configure the client with your credentials
    session = boto3.session.Session()
    client = session.client('s3',
                            region_name='nyc3',
                            endpoint_url='https://nyc3.digitaloceanspaces.com',
                            aws_access_key_id=ACCESS_ID,
                            aws_secret_access_key=SECRET_KEY)
    
    # Define the bucket and object key
    
    filename = os.path.basename(local_file_path)
    object_key = f'{filename}'  # Construct the object key

    # Define the local path to save the file
    

    
    try:
        response=client.upload_file(local_file_path, bucket_name, filename)
        
    except client.exceptions.NoSuchKey:
        return "error: File not found in the bucket"
    except Exception as e:
        return "error: File not found in the bucket"

    # Optional: Send the file directly to the client
    # return send_file(local_file_path, as_attachment=True)

    return "success"

    

def download_file(filename,bucket_name):
    # Configure the client with your credentials
    session = boto3.session.Session()
    client = session.client('s3',
                            region_name='nyc3',
                            endpoint_url='https://nyc3.digitaloceanspaces.com',
                            aws_access_key_id=ACCESS_ID,
                            aws_secret_access_key=SECRET_KEY)
    
    # Define the bucket and object key
    
    object_key = f'{filename}'  # Construct the object key

    # Define the local path to save the file
    local_file_path = os.path.join('downloads', filename)

# Check if the 'downloads' directory exists, create it if not
    if not os.path.exists(os.path.dirname(local_file_path)):
        os.makedirs(os.path.dirname(local_file_path))

    # Download the file from the bucket
    try:
        client.download_file(bucket_name, object_key, local_file_path)
    except client.exceptions.NoSuchKey:
        return "file not in buecket"
    except Exception as e:
        return "exception"

    # Optional: Send the file directly to the client
    # return send_file(local_file_path, as_attachment=True)

    return "success"
    








def get_status(audio_id):
    # Retrieve the task status using the unique ID
    print(audio_id)
    status_info = task_status_tracker.get(audio_id, {"status": "Unknown ID", "percentage": 0})
    return "status"

processed_audio_storage = {}

def api_convert_voice(filename,spk_id1,unique_id):
    acquired = request_semaphore.acquire(blocking=False)
    
    if not acquired:
        return "error in lock"
    #task_status_tracker[unique_id] = {"status": "Starting", "percentage": 0}
    try:
    
        #if session.get('submitted'):
        #    return jsonify({"error": "Form already submitted"}), 400

        # Process the form here...
        # Set the flag indicating the form has been submitted
        #session['submitted'] = True
        
        spk_id = spk_id1+'.pth'
        print("speaker id path=",spk_id)
        voice_transform = 0
        local_file_path = os.path.join('downloads', filename)
        # The file part
        
        file_size = os.path.getsize(local_file_path)
        if file_size > 10 * 1024 * 1024:  # 10 MB limit
            return json.dumps({"error": "File size exceeds 10 MB"}), 400

        content_type_format_map = {
            '.mp3': 'mp3',
            '.wav': 'wav',
            '.mp4': 'mp4',
            '.m4a': 'mp4',
        }
        _, file_extension = os.path.splitext(local_file_path)
        audio_format = content_type_format_map.get(file_extension.lower(), 'mp3') 
        # Default to 'mp3' if content type is unknown (or adjust as needed)
        #audio_format = content_type_format_map.get(file.content_type, 'mp3')

        # Convert the uploaded file to an audio segment
        audio = AudioSegment.from_file(local_file_path, format=audio_format)
        
        
        
    
    # Calculate audio length in minutes
        audio_length_minutes = len(audio) / 60000.0  # pydub returns length in milliseconds
    
        if audio_length_minutes > 5:
            return json.dumps({"error": "Audio length exceeds 5 minutes"}), 400   
            
        #created_files = []
        # Save the file to a temporary path
        #unique_id = str(uuid.uuid4())
        print(unique_id)
        base_filename = os.path.basename(local_file_path)
        
        filename = werkzeug.utils.secure_filename(base_filename)
        input_audio_path = os.path.join(tmp, f"{spk_id}_input_audio_{unique_id}.{filename.split('.')[-1]}")
        #file.save(input_audio_path)
        os.rename(local_file_path, input_audio_path)
        
        #created_files.append(input_audio_path)
        
        #split audio
        task_status_tracker[unique_id] = {"status": "Processing: Step 1", "percentage": 30}

        cut_vocal_and_inst(input_audio_path,spk_id,unique_id)
        print("audio splitting performed")
        vocal_path = f"output/{spk_id}_{unique_id}/{split_model}/{spk_id}_input_audio_{unique_id}/vocals.wav"
        inst = f"output/{spk_id}_{unique_id}/{split_model}/{spk_id}_input_audio_{unique_id}/no_vocals.wav"
        print("*****before making call to convert ", unique_id)
        #task_status_tracker[unique_id] = "Processing: Step 2"
        #output_queue = SimpleQueue()
        #ctx = get_context('spawn')
        #output_queue = ctx.Queue()
    # Create and start the process
        output_path=worker(spk_id, vocal_path, voice_transform, unique_id)
        #p = ctx.Process(target=worker, args=(spk_id, vocal_path, voice_transform, unique_id, output_queue,))
        #p.start()
    
    # Wait for the process to finish and get the result
        #p.join()
        #print("*******waiting for process to complete ")
        
        #output_path = output_queue.get()
        
        task_status_tracker[unique_id] = {"status": "Processing: Step 2", "percentage": 80}
        #if isinstance(output_path, Exception):
         #   print("Exception in worker:", output_path)
        #else:
         #   print("output path of converted voice", output_path)
        #output_path = convert_voice(spk_id, vocal_path, voice_transform,unique_id)
        output_path1= combine_vocal_and_inst(output_path,inst,unique_id)
        
        processed_audio_storage[unique_id] = output_path1
        #session['processed_audio_id'] = unique_id 
        task_status_tracker[unique_id] = {"status": "Finalizing", "percentage": 100}
        print(output_path1)
        upload_file(output_path1,"sing")
        print("file uploaded")
        #created_files.extend([vocal_path, inst, output_path])
        task_status_tracker[unique_id]["status"] = "Completed"
        
    finally:
        request_semaphore.release()
    #if os.path.exists(output_path1):
        
    #    return send_file(output_path1, as_attachment=True)
    #else:
    #    return jsonify({"error": "File not found."}), 404

def convert_voice_thread_safe(spk_id, vocal_path, voice_transform, unique_id):
    with convert_voice_lock:
        return convert_voice(spk_id, vocal_path, voice_transform, unique_id)



def get_vc_safe(sid, to_return_protect0):
    with convert_voice_lock:
        return get_vc(sid, to_return_protect0)




def worker(spk_id, input_audio_path, voice_transform, unique_id):
    try:
        output_audio_path = convert_voice(spk_id, input_audio_path, voice_transform, unique_id)
        print("output in worker for audio file", output_audio_path)
        #output_queue.put(output_audio_path)
        return output_audio_path
    except Exception as e:
        print("exception in adding to queue")
        return "error in converting voice"
        

def convert_voice(spk_id, input_audio_path, voice_transform,unique_id):
    get_vc(spk_id,0.5)
    print("*****before makinf call to vc ", unique_id)

   
    output_audio_path = vc_single(
        sid=0,
        input_audio_path=input_audio_path,
        f0_up_key=voice_transform,  # Assuming voice_transform corresponds to f0_up_key
        f0_file=None ,
        f0_method="rmvpe",
        file_index=spk_id,  # Assuming file_index_path corresponds to file_index
        index_rate=0.75,
        filter_radius=3,
        resample_sr=0,
        rms_mix_rate=0.25,
        protect=0.33,  # Adjusted from protect_rate to protect to match the function signature,
        unique_id=unique_id
    )
    print(output_audio_path)
    return output_audio_path

def cut_vocal_and_inst(audio_path,spk_id,unique_id):
    
    vocal_path = "output/result/audio.wav"
    os.makedirs("output/result", exist_ok=True)
    #wavfile.write(vocal_path, audio_data[0], audio_data[1])
    #logs.append("Starting the audio splitting process...")
    #yield "\n".join(logs), None, None
    print("before executing splitter")
    command = f"demucs --two-stems=vocals -n {split_model} {audio_path} -o output/{spk_id}_{unique_id}"
    env = os.environ.copy()

# Add or modify the environment variable for this subprocess
    env["CUDA_VISIBLE_DEVICES"] = "0"
    
    
    
    #result = subprocess.Popen(command.split(), stdout=subprocess.PIPE, text=True)
    result = subprocess.run(command.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
    if result.returncode != 0:
        print("Demucs process failed:", result.stderr)
    else:
        print("Demucs process completed successfully.")
    print("after executing splitter")
    #for line in result.stdout:
    #    logs.append(line)
    #    yield "\n".join(logs), None, None
    
    print(result.stdout)
    vocal = f"output/{split_model}/{spk_id}_input_audio/vocals.wav"
    inst = f"output/{split_model}/{spk_id}_input_audio/no_vocals.wav"
    #logs.append("Audio splitting complete.")


def combine_vocal_and_inst(vocal_path, inst_path, output_path):
    
    vocal_volume=1
    inst_volume=1
    os.makedirs("output/result", exist_ok=True)
    # Assuming vocal_path and inst_path are now directly passed as arguments
    output_path = f"output/result/{output_path}.mp3"
    #command = f'ffmpeg -y -i "{inst_path}" -i "{vocal_path}" -filter_complex [0:a]volume={inst_volume}[i];[1:a]volume={vocal_volume}[v];[i][v]amix=inputs=2:duration=longest[a] -map [a] -b:a 320k -c:a libmp3lame "{output_path}"'
    #command=f'ffmpeg -y -i "{inst_path}" -i "{vocal_path}" -filter_complex "amix=inputs=2:duration=longest" -b:a 320k -c:a libmp3lame "{output_path}"'
    # Load the audio files
    print(vocal_path)
    print(inst_path)
    vocal = AudioSegment.from_file(vocal_path)
    instrumental = AudioSegment.from_file(inst_path)

# Overlay the vocal track on top of the instrumental track
    combined = vocal.overlay(instrumental)

# Export the result
    combined.export(output_path, format="mp3")

    #result = subprocess.run(command.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    return output_path



def vc_single(
    sid,
    input_audio_path,    
    f0_up_key,
    f0_file,
    f0_method,
    file_index,
    index_rate,
    filter_radius,
    resample_sr,
    rms_mix_rate,
    protect,
    unique_id
):  # spk_item, input_audio0, vc_transform0,f0_file,f0method0
    global tgt_sr, net_g, vc, hubert_model, version, cpt
    print("***** in vc ", unique_id)

    try:
        logs = []
        print(f"Converting...")
        
        audio, sr = librosa.load(input_audio_path, sr=16000, mono=True)
        print(f"found audio ")
        f0_up_key = int(f0_up_key)
        times = [0, 0, 0]
        if hubert_model == None:
            load_hubert()
        print("loaded hubert")
        if_f0 = 1
        audio_opt = vc.pipeline(
            hubert_model,
            net_g,
            0,
            audio,
            input_audio_path,
            times,
            f0_up_key,
            f0_method,
            file_index,
            # file_big_npy,
            index_rate,
            if_f0,
            filter_radius,
            tgt_sr,
            resample_sr,
            rms_mix_rate,
            version,
            protect,
            f0_file=f0_file
        )
        
    
    # Get the current thread's name or ID
        
        
     
        if resample_sr >= 16000 and tgt_sr != resample_sr:
            tgt_sr = resample_sr
        index_info = (
            "Using index:%s." % file_index
            if os.path.exists(file_index)
            else "Index not used."
        )
        
        print("writing to FS")
        #output_file_path = os.path.join("output", f"converted_audio_{sid}.wav")  # Adjust path as needed
        # Assuming 'unique_id' is passed to convert_voice function along with 'sid'
        print("***** before writing to file outout ", unique_id)
        output_file_path = os.path.join("output", f"converted_audio_{sid}_{unique_id}.wav")  # Adjust path as needed

        print("******* output file path ",output_file_path)
        os.makedirs(os.path.dirname(output_file_path), exist_ok=True)  # Create the output directory if it doesn't exist
        print("create dir")
        # Save the audio file using the target sampling rate
        sf.write(output_file_path, audio_opt, tgt_sr)
        
        print("wrote to FS")

        # Return the path to the saved file along with any other information
        
        return output_file_path
           
            
    except:
        info = traceback.format_exc()
        
        return info, (None, None)




def get_vc(sid, to_return_protect0):
    global n_spk, tgt_sr, net_g, vc, cpt, version, weights_index
    if sid == "" or sid == []:
        global hubert_model
        if hubert_model is not None:  # 考虑到轮询, 需要加个判断看是否 sid 是由有模型切换到无模型的
            print("clean_empty_cache")
            del net_g, n_spk, vc, hubert_model, tgt_sr  # ,cpt
            hubert_model = net_g = n_spk = vc = hubert_model = tgt_sr = None
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
            ###楼下不这么折腾清理不干净
            if_f0 = cpt[sid].get("f0", 1)
            version = cpt[sid].get("version", "v1")
            if version == "v1":
                if if_f0 == 1:
                    net_g = SynthesizerTrnMs256NSFsid(
                        *cpt[sid]["config"], is_half=config.is_half
                    )
                else:
                    net_g = SynthesizerTrnMs256NSFsid_nono(*cpt[sid]["config"])
            elif version == "v2":
                if if_f0 == 1:
                    net_g = SynthesizerTrnMs768NSFsid(
                        *cpt[sid]["config"], is_half=config.is_half
                    )
                else:
                    net_g = SynthesizerTrnMs768NSFsid_nono(*cpt[sid]["config"])
            del net_g, cpt
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
            cpt = None
        return (
            gr.Slider.update(maximum=2333, visible=False),
            gr.Slider.update(visible=True),
            gr.Dropdown.update(choices=sorted(weights_index), value=""),
            gr.Markdown.update(value="# <center> No model selected")
        )
    print(f"Loading {sid} model...")
    selected_model = sid[:-4]
    cpt[sid] = torch.load(os.path.join(weight_root, sid), map_location="cpu")
    tgt_sr = cpt[sid]["config"][-1]
    cpt[sid]["config"][-3] = cpt[sid]["weight"]["emb_g.weight"].shape[0]
    if_f0 = cpt[sid].get("f0", 1)
    if if_f0 == 0:
        to_return_protect0 = {
            "visible": False,
            "value": 0.5,
            "__type__": "update",
        }
    else:
        to_return_protect0 = {
            "visible": True,
            "value": to_return_protect0,
            "__type__": "update",
        }
    version = cpt[sid].get("version", "v1")
    if version == "v1":
        if if_f0 == 1:
            net_g = SynthesizerTrnMs256NSFsid(*cpt[sid]["config"], is_half=config.is_half)
        else:
            net_g = SynthesizerTrnMs256NSFsid_nono(*cpt[sid]["config"])
    elif version == "v2":
        if if_f0 == 1:
            net_g = SynthesizerTrnMs768NSFsid(*cpt[sid]["config"], is_half=config.is_half)
        else:
            net_g = SynthesizerTrnMs768NSFsid_nono(*cpt[sid]["config"])
    del net_g.enc_q
    print(net_g.load_state_dict(cpt[sid]["weight"], strict=False))
    net_g.eval().to(config.device)
    if config.is_half:
        net_g = net_g.half()
    else:
        net_g = net_g.float()
    vc = VC(tgt_sr, config)
    n_spk = cpt[sid]["config"][-3]
    weights_index = []
    for _, _, index_files in os.walk(index_root):
        for file in index_files:
            if file.endswith('.index') and "trained" not in file:
                weights_index.append(os.path.join(index_root, file))
    if weights_index == []:
        selected_index = gr.Dropdown.update(value="")
    else:
        selected_index = gr.Dropdown.update(value=weights_index[0])
    for index, model_index in enumerate(weights_index):
        if selected_model in model_index:
            selected_index = gr.Dropdown.update(value=weights_index[index])
            break
    return (
        gr.Slider.update(maximum=n_spk, visible=True),
        to_return_protect0,
        selected_index,
        gr.Markdown.update(
            f'## <center> {selected_model}\n'+
            f'### <center> RVC {version} Model'
        )
    )



def handler(job):
    job_input = job["input"]  # Access the input from the request.
    filename=job_input["filename"]
    spk_id=job_input["spk_id"]
    unique_id=job_input["unique_id"]
    download_file(filename,"sing")
    api_convert_voice(filename,spk_id,unique_id)
    # Add your custom code here.
    return "Your job results"    


runpod.serverless.start({"handler": handler})  # Required.